Methods	Between June 2002 and May 2004, 60 patients undergoing unilateral total knee arthroplasty (TKA) were enrolled in this prospective randomised trial. Randomisation was by sealed opaque envelopes which were mixed by independent personnel and consecutively assigned a case number from 1 to 60. All surgeries were performed by specialists of the joint and reconstruction team using an identical surgical approach and technique. Near the end of the operation the corresponding envelope was opened and the surgeon was informed at the time of drain insertion to achieve a single-blind effect
Participants	60 patients undergoing unilateral total knee arthroplasty (TKA) were randomly allocated to one of two groups: Group 1 (Autotransfusion group): n=26; M//F=6//20; mean (range) age = 72 (57-84) years Group 2 (Control group): n=34; M//F=12//22; mean (range) age = 69.4 (55-78) years
Interventions	Group 1: Autotransfusion group (DONOR system) had their blood reinfused from drains using a 40um blood filter between the collection bag and the intravenous site within 6 hours of surgery. All patients had their drains removed on post-operative day 2 or 3. The DONOR system is an integrated, closed system designed for the collection and reinfusion of drained wound blood. It consists of an 800ml chlorinefree, pre-evacuated collection vessel, a vacuum regulator, and a 40um integrated filter for salvaged blood. Group 2: Control group received no post-operative autotransfusion.
Outcomes	Outcomes reported: amount of allogeneic blood transfused, number of patients transfused allogeneic blood, febrile complications, adverse events, blood loss
Notes	Transfusion threshold: allogeneic blood transfusion was given if the haemoglobin level was less than 9.0g/dL
Risk of bias
Bias	Authors’ judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Method used to generate allocation sequences was adequate.
Allocation concealment (selection bias)	High risk	Sealed opaque envelopes were used to conceal treatment allocation
Blinding (performance bias and detection bias) All outcomes	High risk	Single-blind design.